Validating and Understanding the ENSO Simulation in Two Coupled Climate Models

Validating and understanding the ENSO simulation in two coupled climate models

∗ By VASUBANDHU MISRA , LARRY MARX, JAMES L. KINTER III, BEN P. KIRTMAN, ZHICHANG GUO, DUGHONG MIN, MIKE FENNESSY, PAUL A. DIRMEYER, RAMESHAN KALLUMMAL andDAVID M. STRAUS, Center for Ocean-Land-Atmosphere Studies, Institute of Global Environment and Society, Inc. 4041 Powder Mill Road, Suite 302, Calverton, MD 20705, USA

(Manuscript received 29 July 2006; in ﬁnal form 19 January 2007)

ABSTRACT A newly developed Atmospheric General Circulation Model (AGCM) at T62 spectral truncation with 28 terrain- following (σ = p ) levels coupled to the Modular Ocean Model version 3.0 (MOM3.0) is evaluated for its simulation ps of El Niño and the Southern Oscillation (ENSO). It is also compared with an older version of the AGCM coupled to the same ocean model. A dozen features of ENSO are validated. These characteristics of ENSO highlight its influence on global climate at seasonal to interannual scales. The major improvements of the ENSO simulation from this new coupled climate model are the seasonal phase locking of the ENSO variability to a realistic annual cycle of the eastern equatorial Pacific Ocean, the duration of the ENSO events and its evolution that is comparable to the ocean data assimilation. The two apparent drawbacks of this new model are its relatively weak ENSO variability and the presence of erroneous split ITCZ. The improvement of the ENSO simulation in the new coupled model is attributed to realistic thermocline variability and wind stress simulation.

1. Introduction In this study, we first validate the simulation of ENSO in two different versions of a CGCM. The CGCM used in this study El Niño and the Southern Oscillation (ENSO) variability is one has as its atmospheric component the Center for Ocean-Land- of the largest sources of interannual variability driven by cou- Atmosphere Studies (COLA) Atmospheric General Circulation pled climate processes (Philander, 1983, 1990) that has ramifi- Model (AGCM) Version2.2 (V2.2) and the recently updated cations on the global climate (Rasmusson and Carpenter, 1982; COLA AGCM Version3.2 (V3.2). The AGCMs incorporate a Ropelewski and Halpert, 1987; Trenberth et al., 1998; Goddard model of the land surface as described in the following sec- and Dilley, 2005). Warm (cold) ENSO episodes are character- tion. The ocean model is the Modular Ocean Model Version 3.0 ized by increased (decreased) SST in the central and the eastern (MOM3; Pacanowski and Griffies, 1998). The realism of ENSO equatorial Pacific Ocean with a broad-band periodicity of 2 to simulations has been evaluated in several studies (Neelin et al., 7 yr. This variation in SST has concomitant changes in the sur- 1992; Mechoso et al., 1995; Latif et al., 2001; van Oldenborgh face wind stress, sea level pressure and ocean heat content as well et al., 2005; Capotondi et al., 2006; Guilyiardi, 2006; Joseph and as the teleconnection patterns of precipitation, surface tempera- Nigam, 2006; Rao and Sperber, 2006). ture, and tropospheric temperature that are manifestations of the In this paper, we have highlighted 12 features related to ENSO complex interactions of the coupled oceanÐlandÐatmosphere dy- suitable for verifying the credibility of a global coupled model namical system. The variations in the tropical Pacific are primar- simulation. We argue that these features compose a necessary but ily due to coupled oceanÐatmosphere modes (Philander, 1983; not sufficient condition for a successful simulation of ENSO. In Suarez and Schopf, 1988; Jin, 1997), while the tropical and extra- addition, we restrict the analysis to these 12 features due to the tropical teleconnections occur primarily through atmospheric constraints of the available observations and our limited under- bridges (Lau and Nath, 1996; Alexander et al., 2002). standing of the phenomenon. The 12 features are as following:

(1) Mean upper ocean and precipitation errors in the tropics. ∗ Corresponding author. (2) Mean annual cycle of the Equatorial Paciﬁc SST and e-mail: [email protected] wind stress. DOI: 10.1111/j.1600-0870.2007.00231.x (3) Temporal Variability of the Ni˜no3 SST.

292 Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 293

(4) Seasonal phase locking of the Niño3 SST variability. simulation between the two coupled models followed by con- (5) ENSO related SST and upper Ocean variability. cluding remarks in Section 6. (6) Duration of ENSO events. (7) Evolution of SST in the equatorial Pacific. 2. Model description (8) Evolution of the subsurface ocean temperatures in the equatorial Pacific. 2.1. Atmospheric general circulation model (9) Relationship of the Niño3 SST variability with the tropical Oceans. A new AGCM has been recently developed at COLA (COLA (10) Relationship between Niño3 SST and wind stress. AGCM V3.2; hereafter V3.2). The previous version of the COLA (11) Relationship between Niño3 SST and precipitation. AGCM (V2.2) which formed the basis for the current version of (12) Mid-latitude atmospheric response to ENSO. the AGCM has been extensively used for coupled climate modelling studies in the past (Kirtman and Shukla, 2002; Kirtman Despite the many intercomparison studies of coupled models, et al., 2002; Schneider, 2002; Kirtman, 2003). However, it should and mechanistic and process studies relating to ENSO, there be noted that V2.2 had some serious flaws in its mean climate as are considerable gaps in our understanding of the phenomenon well as in its ENSO simulation with annual mean errors of SST ◦ that is largely reflected in the relatively poor prediction and in excess of 5 in the tropical Pacific. simulation skill of state of the art CGCMs. Most current coupled V3.2 has a revised subgrid scale physical parametrizations models have an erroneously split intertropical convergence package compared to V2.2 (Table 1). An important aspect of zone (ITCZ), cold bias in the equatorial eastern Pacific, a V3.2 is that the atmospheric component has been tuned to pro- warm bias off the South American and the South African duce a reasonable mean climate that balances the energy at the coasts, a tropical thermocline that is too diffuse, insufficient top of the atmosphere and at the surface to within a couple of − upwelling along the eastern boundaries of continents, interan- Wm 2 when coupled to the ocean model (MOM3). Many cli- nual variability that extends too far to the west and is narrow mate modelling centres develop component models independent and weak along the equator. These deficiencies have been of each other before finally coupling them. The latter approach documented in a workshop on tropical biases in coupled models does not account for the biases that arise as a result of errors (ftp://grads.iges.org/pub/schneider/CTBW05/Previous mtg/ in coupling the components of the climate system and the non- Ping CLIVAR summary.pdf;ftp://grads.iges.org/pub/schneider/ linear coupled error growths arising in the dependent variables CTBW05/Previous mtg/EPIC tropbias Bretherton.2003.pdf). of the component models. Furthermore, by forcing (tuning) the In addition, many models exhibit a strong biennial cycle in the climate of the component models to behave reasonably relative SST variability over the eastern Pacific that is unsupported by to observations without taking into account coupled processes observations. Some of these issues persist in the two coupled can give rise to additional biases in the coupled climate system climate models used in this study as well. It should however (Wu et al., 2006). be noted that these deficiencies act as a limiting barrier on the The tuning in V3.2 involved modulating the vertical profile of predictability of ENSO in the current state-of-the art coupled diabatic heating through the adjustment of the fractional amount climate models. of large-scale and convective precipitation that is evaporated into The details of the coupled model used in this study are outlined the environment, the critical relative humidity to diagnose sat- in the following section followed by a description of the design of uration cloud fraction and the order of a spatial filter used to experiments. The results are presented in Section 4. In Section 5, smooth the inversion clouds (which is required as a result of the we discuss the results to understand the differences in the ENSO spectral truncation) that significantly affected the bias over the

Table 1. The outline of the physics package in the AGCM used in the study (V3.2) and its comparison with the previous version (V2.2)

Process V3.2 V2.2

Deep convection Relaxed Arakawa Schubert Relaxed Arakawa Schubert scheme (Bacmeister et al., 2000) scheme (Moorthi and Suarez, 1992) Longwave radiation Collins et al. (2002) Harshvardhan et al. (1987) Boundary layer non-local (Hong and Pan, 1996) level 2.0 closure (Mellor and Yamada, 1982) Land surface process Xue et al. (1991, 1996); Dirmeyer and Zeng (1999) Xue et al. (1991, 1996) Shallow convection Tiedtke (1984) Tiedtke (1984) Shortwave radiation Briegleb (1992) Briegleb (1992) Diagnostic cloud fraction and optical properties Kiehl et al. (1998) Kiehl et al. (1998)

Tellus 59A (2007), 3 294 V. MISRA ET AL.

stratocumulus region. It was found that to understand the efficacy and allowed for variation of the latent heat of phase change with of a given change in this coupled model on the mean state and temperature following Bohren and Albrecht (1998) in V3.2. on the seasonal to interannual scales of variability, multidecadal integrations (over 50 yr) were required for the upper ocean to 2.2. The ocean model equilibrate. V3.2 is now run at exactly the same resolution as the NCEP The ocean model is version 3 of the Geophysical Fluid Dynam- reanalysis model (T62L28; Kalnay et al., 1996) with identical to- ics Laboratory MOM (Pacanowski and Griffies, 1998). It uses a pography. This was primarily done to conduct seasonal hindcasts finite difference treatment of the primitive equations of motion with this AGCM using initial conditions taken from the NCEP using the Boussinesq and hydrostatic approximations in spheri- reanalysis avoiding interpolation problems. The vertical coordi- cal coordinates. The domain is that of the world ocean between nate system is the terrain following sigma coordinate. V2.2 is run 74◦S and 65◦N. The coastline and bottom topography are real- at T42 horizontal spectral truncation with 18 terrain following istic except that ocean depths less than 100 m are set to 100 m sigma levels. The dynamical core is based on CCM3.6.6 (Kiehl and the maximum depth is set to 6000 m. The artificial high- et al., 1998) as in V2.2. Here, all prognostic variables except the latitude meridional boundaries are impermeable and insulating. moisture variable are treated spectrally. Moisture is advected by The zonal resolution is 1.5◦. The meridional grid spacing is 0.5◦ the Semi-Lagrangian scheme. between 10◦S and 10◦N, gradually increasing to 1.5◦ at 30◦N and The outline of the physics of V3.2 is presented in Table 1 and 30◦S and fixed at 1.5◦ in the extratropics. There are 25 levels in compared with V2.2 of the model. For the planetary boundary the vertical, with 17 levels in the upper 450 m. The vertical mix- layer (PBL) we have adopted the non-local scheme of Hong and ing scheme is the non-local K profile parametrization of Large Pan (1996). Local-K theory, on which the planetary boundary et al. (1994). The horizontal mixing of tracers and momentum is layer scheme of V2.2 is based, parametrizes turbulent mixing Laplacian. The momentum mixing uses the space-time depen- with an eddy diffusivity based on local gradients of wind and dent scheme of Smagorinsky (1963) and the tracer mixing uses temperature. This has the potential to fail in unstable bound- Redi (1982) diffusion along with Gent and McWilliams (1990) ary layers because the influence of large eddy transports is not quasi-adiabatic stirring. accounted for (Troen and Mahrt, 1986; Holtslag and Moeng, 1991). The non-local scheme following Hong and Pan (1996) 3. Design of experiments parametrizes the counter-gradient transport effected through the large-scale eddies. The long wave scheme in V3.2 follows that For the new coupled model simulation (V3.2-MOM3; hereafter of Collins et al. (2002) which was developed from water vapor NEW), the ocean model from the initial state of rest with in- line and continuum treatments that use a line-by-line radiative tial conditions of temperature and salinity of Levitus (1982) was transfer model (GENLEN2) which is more accurate than the forced with climatological wind stress derived from special sen- broad-band absorptance method used in V2.2. Although the deep sor microwave imager (SSMI) for a period of 100 yr. There- convection is still the Relaxed Arakawa Schubert (RAS) scheme after, the model was integrated in the coupled system with earlier in V3.2, the determination of the fraction of the detrained cloud versions (V3.0 and V3.1) of the AGCM for a period of 80 yr. liquid water is done by means of an explicit budget equation The coupled model (V3.2 + MOM3) was further integrated for in the cloud model (Bacmeister et al., 2000) in V3.2 instead of 95 yr from such an initial state of the ocean. The results shown an empirical profile derived from GATE (Moorthi and Suarez, are computed using the last 70 yr of the integration discarding 1992) in V2.2. The land surface model follows the formulation the first 25 yr as a spin-up period. of the Simplified Simple Biosphere (SSiB) scheme (Dirmeyer The coupled model (V2.2-MOM3; hereafter OLD) integration and Zeng, 1999) based on Xue et al. (1991, 1996). The version details follows from Kirtman et al. (2002). It is a 70 yr coupled of SSiB in V3.2 was used in the second Global Soil Wetness run with the last 40 yr of the integration being shown here. The Project (International GEWEX Project Office, 2002). Among ocean initial state for this coupled run was spun up from an initial the most significant differences from the earlier version (V2.2) state of rest with initial conditions of temperature and salinity is that the soil temperature and wetness are predicted in six lay- of Levitus (1982) forced with climatological wind stress from ers (instead of three), with four embedded within the root zone. SSMI for a period of 100 yr. This provides a more consistent and realistic simulation of the decrease of transpiration with increasing soil moisture stress, 4. Result preventing so-called ‘stomatal suicide’. The fourth order horizontal diffusion coefficient was reduced by 2 orders of magnitude In discussing and validating the model results we use the Hadley in V3.2 relative to V2.2. Second order horizontal diffusion was Center Global Sea Ice and Sea Surface Temperature (HADISST, introduced in the top two layers of the V3.2 model to achieve nu- Version 1.1; Rayner et al., 2002; in the text interchangably used merical stability. Additionally, we have implemented a uniform with observations) for monthly mean observed SST available calculation of saturation vapor pressure following Marx (2002) at 1◦ resolution, NCEP-NCAR reanalysis (Kalnay et al., 1996)

Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 295

made available on a 2.5◦ × 2.5◦ latitudeÐlongitude grid for atmo- SST between the simulation and the observations) in the trop- spheric variables, gridded precipitation on 2.5◦ × 2.5◦ latitudeÐ ical oceans from the OLD and the NEW coupled runs. The longitude grid from the Climate Prediction Center Merged Anal- OLD (NEW) model has a warm (cold) bias in the equatorial ysis Precipitation (CMAP; Xie and Arkin, 1996) and an ocean Pacific region with a root mean square error of 3.91◦(0.94◦) and data assimilation (ODA) analysis of available ocean observations 3.19◦(1.2◦) over the Niño3 and Niño3.4 regions, respectively. In following Rosati et al. (1997) for subsurface ocean quantities. the OLD model, there was excessive downwelling shortwave at All the results shown are computed from all calendar months un- the surface (Schneider, 2002) that contributed to the warm bias less mentioned otherwise. In computing climatological means, in the equatorial Pacific Ocean. The error in the NEW model is we use 40 (70) yr of the OLD (NEW) model integration, the last typical of the models exhibiting the split-ITCZ problem. A vast 70 yr (1930Ð2000) of HADISST, all 52 yr of the NCEP-NCAR improvement is observed from the OLD to the NEW coupled reanalysis (1948Ð2000), 22 yr of the CMAP data (1979Ð2000), simulation even over the stratus region in the eastern tropical and 19 yr of the ODA (1980Ð1998). For the OLD run, we have Indian, Pacific and the Atlantic Oceans. These errors over the stored only the atmospheric variables for the last 40 yr of the cou- equatorial Ocean and in the eastern Oceans are comparable to pled integration. Therefore, the comparison between the OLD those in many of the IPCC models used in AR-4 (not shown; and the NEW coupled simulations will not include subsurface Covey et al., 2003; Guilyiardi, 2006). variables. The different duration of the analysis periods for the The depiction of this mean error is intuitively important in NEW and OLD simulations does not seem to be an issue. In fact, evaluating the SST anomalies and precipitation anomalies as- the results presented here for the NEW model did not change sig- sociated with ENSO variability. This is further substantiated in nificantly when only 40 yr were used. Figs. 2a and b showing the climatological annual mean precipita- Although in this study we have concentrated on the sea- tion errors from the two simulations. These figures are consistent sonal to interannual variability of the ENSO phenomenon, longer with climatogical annual mean SST errors (Figs. 1a and b), in time variations of ENSO (Gu and Philander, 1995; Wang, 1995; that, for the most part at least in the OLD model the wet (dry) bias Timmermann, 1999) make such validation somewhat uncertain in precipitation follows the warm (cold) bias in SST errors. In the and can limit the understanding of the ENSO phenomenon from NEW model the dry bias over the central and eastern equatorial the available short records of observations. Model errors are Pacific Ocean is straddled by a wet bias in both hemispheres as larger than observational errors, so that inaccuracies in observa- typical of the split ITCZ problem. It is interesting to note that tions do not prevent the identification of model deficiencies. a similar split ITCZ problem exists over the tropical Atlantic Ocean. The cold bias in the equatorial Pacific in the NEW model is accompanied by an easterly bias in the low level winds over 4.1. Mean upper ocean and precipitation errors the equatorial Pacific (not shown). In Figs. 1a and b, we show the mean annual SST errors (com- The mean error in the thermocline depth in the NEW model as puted as the climatological mean difference of the annual mean assessed by the depth of the 20◦C isotherm is shown in Fig. 3a.

Fig. 1. The climatological annual mean errors of SST (in ◦C) from the (a) OLD and the (b) NEW coupled simulation. The outline of the Ni˜no3 (Ni˜no3.4) in solid (thick dashed line) is also shown.

Tellus 59A (2007), 3 296 V. MISRA ET AL.

Fig. 2. Same as Fig. 1 but for precipitation. The units are in mm d−1.

Fig. 3. (a) The annual mean errors of the thermocline depth (in meters) from the NEW model based on ODA (Rosati et al., 1997) over the tropical oceans. The thermocline depth is diagnosed from the depth of the 20◦C isotherm. (b) The thermocline depth along the equatorial Paciﬁc from the ODA (solid line) and the NEW coupled simulation (dashed line).

It is important indicator to gauge the errors in the thermocline relatively deeper thermocline over the tropical Indian Ocean with depth, as the feedback of the subsurface ocean is an important respect to the ODA. The errors in the thermocline depth increase of ENSO variability (Philander, 1983; Suarez and Schopf, 1988; poleward of 5◦ latitude in either hemisphere over the tropical An and Jin, 2001) The model clearly has a relative minimum of oceans. systematic error in the equatorial oceans straddled by the positive The mean thermocline depth in the equatorial Paciﬁc Ocean (negative) errors to the (north) south of the equator in the tropical is shown in Fig. 3b for the NEW model simulation. The Paciﬁc and the Atlantic Oceans. The model tends to simulate a NEW model is able to capture the sharp zonal gradient in the

Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 297

Fig. 4. The climatological annual cycle of SST in the equatorial Pacific Ocean from (a) observations (HADISST) the (b) OLD and (c) NEW coupled model simulation. thermocline depth from the western Pacific where it is relatively the warm SST in the early part of the year is also well simulated deep to the eastern Pacific where it nearly shoals to the surface. by the NEW model. However, the pronounced annual cycle in the This sharp gradient is found to be critical to the onset and phase far western Pacific Ocean simulated by both models is unrealis- locking of the ENSO (An and Jin, 2001; Jin, 1997). tic. In addition, the cold (warm) SSTA in the boreal fall (spring) is smaller (larger) in amplitude in the NEW model relative to the observations. 4.2. Mean annual cycle of the equatorial pacific sst and In Figs. 5 and 6, we show the annual cycle of the zonal and wind stress the meridional wind stress from the NCEP reanalysis, OLD The successful simulation of both the mean annual cycle and the and NEW model simulations. In both of these figures the NEW interannual variability in the tropics has proven to be surprisingly model simulates the annual cycle of the wind stress (both zonal difficult in many CGCMs (Ineson and Davey, 1997). The climate and meridional components) in the equatorial eastern Pacific drift in the mean annual cycle of SST in the tropics is a serious Ocean in a way that conforms well with observations. The OLD flaw in many of the CGCMs (Mechoso et al., 1995) that can have model displays a distinct semi-annual cycle especially in the serious impacts on CGCM seasonal forecasts (Tziperman et al., zonal wind stress (Fig. 5b). In addition, the zonal wind stress 1997; Yang and Anderson, 2000). anomalies in the eastern Pacific in the OLD model is weaker In Figs. 4aÐc, we show the climatological mean annual cycle than in the NEW model and NCEP reanalysis. However, both of the equatorial Pacific SST from HADISST, OLD and the NEW models exhibit stronger wind stress annual cycle over the equa- coupled model simulations, respectively. As depicted in Fig. 4a, torial western Pacific region compared to NCEP reanalysis. The the observations show a distinct annual cycle over the eastern Pa- large anomalies of the meridional wind stress in the OLD model cific with the warmest (coldest) SST occurring in MarchÐApril (Fig. 6b) are farther westward from the eastern boundary con- (AugustÐSeptember). The OLD coupled model has a weaker an- trary to NCEP reanalysis (Fig. 6a) and the NEW model (Fig. 6c). nual cycle in the eastern Pacific with a hint of a semi-annual cy- However, both models exhibit stronger wind stress annual cycle cle. The NEW coupled model in Fig. 4c reproduces the observed over the equatorial western Pacific region compared to NCEP re- mean annual cycle reasonably well. The westward migration of analysis. But the stronger than observed meridional wind stress

Fig. 5. Same as Fig. 4 but for zonal wind stress. The units are in N m−2.

Tellus 59A (2007), 3 298 V. MISRA ET AL.

Fig. 6. Same as Fig. 4 but for meridional wind stress. The units are in N m−2. in both models over central equatorial Pacific reflects the strong broad peak of the observed spectrum. However, compared to meridional (Hadley-type) overturning in the coupled simulations HADISST the sharpness of the spectrum in the NEW and the as a result of the split ITCZ (Nigam et al., 2000). OLD model suggests that ENSO is relatively regular in the coupled simulations. Furthermore, both models exhibit insignificant low frequency (>4 yr) variability. 4.3. Temporal variability of Nino3˜ SST ENSO has a characteristic feature of having a broad peak in the 4.4. Seasonal phase locking of the Nino3˜ SST variability range of 2Ð7 yr. This feature has been often validated in models by examining the sample spectrum of the Niño3 SST (Meehl and The apparent phase locking of ENSO events to the mean annual Arblaster, 1998). In Fig. 7, we show the Niño3 SST spectrum cycle in the eastern equatorial Pacific with a tendency to peak at from both the observations (HADISST) and the coupled model the end of the calendar year is perhaps one of ENSO’s most dis- simulations. HADISST displays a broad peak from 3 to 7 yr. The tinctive characteristics (Rasmusson and Carpenter, 1982). ENSO OLD (NEW) model has a characteristic feature of strong (weak) variability typically peaks in boreal winter and diminishes in bo- bi-ennial cycle. The strongest peak of the Niño3 SST spectrum real spring with relatively weak variability in the boreal summer in the NEW model is around 3 yr which just falls within the and early fall. In Fig. 8, we show the monthly standard deviation of the SST anomalies in the Niño3 region from the observations and the two coupled model simulations. It is apparent from the SST Spectrum over Nino3 Region 2.0 figure that the NEW coupled model in contrast to the OLD model HADISST has a preference for relatively high variability in the winter sea- OLD NEW son, as observed. However, the standard deviations in the NEW 1.5 simulation are much smaller than observed. It should however

1.0 Power

0.5

0.0 5.0 2.5 2.0 1.25 1.0 Period (years) Fig. 7. The sample spectrum of SST over the Niño3 region is shown from observations (HADISST) and the coupled simulations. The dashed line in the figure shows the lower limit of the 95% confidence interval of the observed sample spectrum for 4◦ of freedom according to chi-squared test. The upper limit of the confidence interval is not shown for clarity. Niño3 SST at different lead times. The Fig. 8. The standard deviation of SST over the Niño3.4 region from autocorrelation of first order autoregressive process (in red) is also observations (HADISST; solid line), the OLD (short dashed line) and shown. the NEW (long dashed line) model simulations.

Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 299

Fig. 10. The regression of the Niño3 SST index on the thermocline depth (in meters) in the tropical Oceans from (a) the ODA and (b) the NEW model simulation. Only values significant at the 90% confidence interval according to the t-test are plotted.

Fig. 9. The regression of the Niño3 SST index on tropical SST from ing) in the western (eastern) equatorial Pacific Ocean. SSTA are (a) Observations (HADISST) the (b) OLD and (c) NEW coupled model the manisfestation of an ENSO event while the thermocline depth simulations. Only values significant at the 90% confidence interval anomalies act as precursors and also mark the end of a ENSO according to the t-test are plotted. event (Zelle et al., 2004). In Figs. 10a and b, we have plotted the regression of the Niño3 SST index on the corresponding thermocline depth from the ODA and the NEW coupled simulation, be noted that the peak variability of some ENSO events does not respectively. Both show a dipole like structure between the east- necessarily occur in the preferred boreal winter season. In sum- ern and western equatorial Pacific Ocean that fits the standard mary, the NEW model is an improvement over the OLD model schematic of ENSO (Philander, 2001). Furthermore, the mag- in terms of ENSO’s phase locking to the annual cycle. nitude of variability of the simulated thermocline depth in the eastern Pacific is comparable to the ODA. However, it should be noted that the meridional scale of the thermocline depth anoma- 4.5. ENSO related SST and upper ocean variability lies in the NEW model is smaller than in the ODA. The spatial structure of the anomalous SST and upper ocean In Figs. 11a and b, we show the regression of the Niño3 SST variability plays a significant role on the extra-tropical response index on the thermocline depth at a lag of 8 months (Niño3 SST (Hoskins and Karoly, 1981; Kumar and Hoerling, 1998; Chen, index lagging) from the ODA and the NEW model simulation, 2004). Regressions of the Niño3 SST index on tropical Pacific respectively. Meinen and McPhaden (2000) have highlighted the SST from the observations and the OLD and the NEW coupled importance of this variability which has anomalies of the same model simulations are shown in Figs. 9aÐc, respectively. The sign along the equatorial Pacific that describes the variations of NEW model simulation reproduces the observed SST variabil- the zonally averaged thermocline depth on the SST variations. ity over the eastern equatorial Pacific comparatively well. But Following the ‘recharge oscillator’ paradigm Jin (1997), the SST the NEW model has a tendency to produce too much variabil- anomaly in the eastern equatorial Pacific is approximately in ity in the western Pacific. The OLD model displays variabil- quadrature with the phases of maximum variations of zonally ity that is spatially confined to the eastern Pacific which does averaged thermocline depth. Therefore, this pattern of the ther- not extend into central Pacific Ocean. In contrast, the NEW mocline depth acts as a precursor to the Niño3 SST variations model produces excessive variability over the western Pacific (Meinen and McPhaden, 2000). Here (Fig. 11) we identify this Ocean. Furthermore, the NEW model and the OLD model have variation at a lag of 8 months in the ODA and the NEW model a tendency to produce too little variability along the coastline of simulation. It is seen that the variations of the thermocline depth Peru. are much smaller than in Fig. 10 which is consistent with the The cycle of interactions involved in ENSO include changes in diagnosis of Meinen and McPhaden (2000) and Capotondi et al. the thermocline depth (Philander, 1983) with shoaling (deepen- (2006) who identified this variability as the second EOF mode

Tellus 59A (2007), 3 300 V. MISRA ET AL.

months. The change in sign of the autocorrelation curve signiﬁes the connectivity to the opposite phase of variability (Joseph and Nigam, 2006).

4.7. Evolution of SST in the equatorial pacific Many coupled models have a tendency of being in a state of perpetual ENSO, that is, one event of ENSO would lead to the other. This leads to the absence of neutral events in such models. This is often seen when the duration of the ENSO events are comparable to half the time period of the spectral peak. In Figs. 13aÐc, we display the lag/lead regression of the Niño3 SST with the equatorial Pacific SST over a 42 month interval from both the observations and the two coupled model simulations, respectively. Positive lag indicates that the Pacific SST leads the standardized Niño3 SST index. The observations show an asymmetry in the evolution of equatorial SST with the SST Fig. 11. Same as Fig. 10 but with Niño3 SST index lagging the anomalies being larger and slightly shorter in one phase of the thermocline depth by 8 months. evolution relative to the other. The OLD model has a strong biennial cycle with duration of ENSO events typically around Autocorrelation of Nino3 SST 1.0 10 months (Fig. 12) that puts it in a perpetual ENSO mode. The

Observation evolution of equatorial Paciﬁc SST is rather well captured by NEW OLD the NEW coupled model besides the east-west coherent structure of the SST anomalies. However, the NEW coupled model extends its warm SST anomalies too far to the west relative to 0.5 1 the observations. e =0.368

4.8. Evolution of the subsurface ocean temperatures in

Autocorrelation the equatorial pacific 0.0 In Fig. 14, we show the regression of the Niño3 SST index on the equatorial subsurface ocean temperatures at 12, 8, 4, 0 and -4 months lead from the ODA and the NEW model simulation. In this figure the standardized Niño3 SST index lags the equatorial Ð0.5 Ð18 Ð12 Ð6 0 6 12 18 Pacific subsurface ocean temperature anomalies at positive lags. lag/lead months In order to make the discussion more lucid in reference to this Fig. 12. The autocorrelation of the Niño3 SST index from observation figure we make an assumption that negative (positive) values (HADISST) and the coupled simulation. Horizontal line is drawn at correspond to cold (warm) Niño3 SST index events. It is apparent −1 = e 0.368 to estimate the duration of the event during one phase of from the figure that the upper ocean anomalies before, during the ENSO cycle. and after the evolution of the warm ENSO event, in the coupled model compare well with the ODA. At 12 months preceding explaining about one-fifth of the total variance. The NEW model the warm ENSO event (Figs. 14a and f) the warm anomalies simulation shows a close resemblance to ODA in this feature. from the west Pacific are shown to move eastward as the cold upper anomaly in the east Pacific recedes. Further, at 8 months (Figs. 14b and g) lead to the ENSO event, the anomalies gain 4.6. Duration of ENSO events magnitude and the eastward tilt with decreasing depth becomes Following the methodology proposed in Joseph and Nigam more prominent. At 4 months prior to the ENSO event (Figs. 14c (2006) we show in Fig. 12 the autocorrelation of the Niño3 SST at and h), the warm anomalies are established in the east Pacific various leads and lags. The width of the autocorrelation curve at while cold anomalies appear in the ODA in the West Pacific. its intersection with e−1(decorrelation values; =0.368) line pro- At zero lag (Figs. 14d and i) the anomalies in the east and west vides an estimate of the duration of the event in any one phase Pacific become larger and extend deeper in the upper ocean. (warm or cold). The width in the observations is about 14 months At 4 months (Figs. 14e and j) past the ENSO event the warm while in the NEW (OLD) model simulation it is around 13 (10) anomalies in the east Pacific begin to retreat in the east Pacific

Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 301

Fig. 13. Lead/lag Niño3 SST index regressions on equatorial Pacific SST fro m (a) Observations (HADISST) (b) the OLD and (c) the NEW coupled model simulations. At positive lags the Niño3 SST index lags the equatorial Pacific SST. Only values significant at the 90% confidence interval according to the t-test are plotted.

Fig. 14. The regression of Niño3 SST index on the equatorial subsurface ocean temperatures from the ocean data assimilation (Rosati et al., 1997) and the NEW coupled model simulation at (a), (f) 12 months lag, (b), (g) 8 months lag (c), (h) 4 months lag, (d), (i) 0 months lag and (e), (j) 4 months lead. At positive lags Niño3 SST index lags the subsurface ocean temperature anomalies. Only values significant at the 90% confidence interval according to the t-test are plotted. while the cold anomalies from the west Pacific propogate further Oceans is comparable to that due to the remote ENSO forc- eastward. Despite the reasonable simulations of the subsurface ing (Chang et al., 2003; Krishnamurthy and Kirtman, 2003). It ocean temperature anomalies, a glaring disparity in the NEW should be noted that these correlations over the Indian and the model simulation is its weaker amplitude of the variability at all Atlantic Oceans are seasonally dependent and involve lead-lag lag/lead times. relationship with the Niño3 SST variability. However, these rela- tioships are so dominant that they are reflected when all calendar months are taken into consideration as shown below. 4.9. Relationship of the Nino3˜ SST variability with the In Fig. 15, we show the contemporaneous correlation of the tropical oceans Niño3 SST index with global tropical SST. The horse-shoe pat- This ENSO metric is extremely important to validate given that tern of the correlation in the tropical Pacific that is evident in the the intrinsic variability in the tropical Atlantic and the Indian observations is simulated by the NEW model simulation while

Tellus 59A (2007), 3 302 V. MISRA ET AL.

Fig. 15. Contemporaneous correlations of Niño3 SST with global tropical oceans from (a) Observations (HADISST), (b) the OLD and (c) the NEW coupled simulations. Only values significant at the 10% significance level according to the t-test are shown.

Fig. 16. The regression of the Niño3 SST index on the zonal wind it is not captured in the OLD simulation. In the tropical Atlantic, stress anomalies from the (a) NCEP reanalysis, (b) the OLD and (c) the the correlations are weaker but significant verifiable positive cor- NEW coupled simulation. Only values significant at the 90% relations appear in the Carribbean Sea in both simulations. These confidence interval according to the t-test are plotted. The units are in −2 correlations over the northern tropical Atlantic region peak in the dynescm . boreal spring following the ENSO peak in the preceding winter (Enfield and Mayer, 1997; Huang, 2004). 4.10. Relationship between Nino3˜ SST and wind stress However, the OLD model is able to capture the positive correlation albeit stronger than observations over the South trop- The relationship between Niño3 SST and wind stress represents ical Atlantic Ocean which is completely missed by the NEW the coupled airÐsea interaction of ENSO. The Bjerknes feed- model. This variability over the southern Atlantic Ocean forced back mechanism (Bjerknes, 1969) which relates tropical ocean by ENSO also appears in the Boreal spring but is found to be adjustment (involving deepening of thermocline and weaken- much weaker in observations than that over the North tropical ing of equatorial upwelling) in response to modulations in the Atlantic Ocean (Huang, 2004). low level trade winds (development of westerly wind anomaly) In the western Indian Ocean, the positive correlations are well that follows from the shift of the Walker circulation eastward, (poorly) represented by the NEW (OLD) coupled model. How- is reflected in this relationship. Besides, as shown in Kirtman ever, the negative correlations in the eastern Indian Ocean ap- (1997), the meridional structure of the wind stress anomaly has pear unrealistic in the NEW model. Since the ENSO variability a bearing on the ENSO period as a result of role played by the is relatively weak in the NEW model, the dominant basin wide off equatorial Rossby waves. positive correlations of the tropical Indian Ocean with Niño3 In Figs. 16aÐc, we show the regression of the Niño3 SST index SST variability is not well simulated. Another problem that the on the zonal wind stress anomalies from the NCEP reanalysis, NEW model displays in this figure is the increased variability the OLD and the NEW model simulations, respectively. The re- over the western Pacific Ocean. Some of the other IPCC (AR4) analysis clearly shows a coherent westerly wind stress anomaly models also suffer with such a bias (Joseph and Nigam, 2006). over the warm pool region in the western Pacific that is stradled

Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 303

The NEW coupled model simulation in Fig. 17c shows a verifiable modulation of precipitation over the central and eastern equatorial Pacific Ocean. The rainfall variability over the tropical north Atlantic is simulated albeit weakly. The split ITCZ phenomenon in the tropical Pacific is clearly seen in the NEW model simulation with the precipitation anomalies too meridion- ally confined suggesting a strong Hadley-like circulation. The figure also shows a general agreement between the simulation and the observations in the Indo-Pacific sector, southern Indian Ocean, and in the south Pacific convergence zone. However, the warm pool region shows enhanced rainfall anomalies in the simulation (that is unsupported by observations), which is consistent with the increased SST variance (cf. Fig. 9c) that extends too far to the west.

4.12. Mid-latitude atmospheric response to ENSO The mid-latitude response to anomalous tropical heating due to ENSO is illustrated from the geopotential height anomalies of the mid-upper troposphere. Horel and Wallace (1981) showed that the equatorial convection forces a train of stationary Rossby waves that arc across the Pacific-North American on a great- circle route. In Figs. 18aÐc the regression of Niño3 SST index on 200 hPa Fig. 17. The regression of the Niño3 SST index on precipitation from geopotential height anomalies for DecemberÐJanuaryÐFebruary (a) Observations (SST from HADISST and precipitation from CMAP), (DJF) are shown from the NCEP reanalysis and the OLD and (b) the OLD and (c) the NEW coupled simulations. Only values the NEW model simulations, respectively. The NEW model sim- significant at the 90% confidence interval according to the t-test are plotted. The units are in mm d−1. ulation roughly resembles the SST forced pattern illustrated in Straus and Shukla (2000; cf. their Fig. 18a) that has broad centres around 45◦N between 140◦W and 160◦E, between (and just west by anomalies of the opposite sign in both hemispheres. This of) the Great Lakes and Hudson Bay and over the Gulf of Mex- is reasonably well captured by the NEW model simulation in ico (near 30◦N). The rather zonally symmetric height response Fig. 16c except that the wind stress anomaly in the warm pool of the OLD model (18b) is contrary to the NCEP reanalysis in region is narrower than the reanalysis and the anomalies off the 18a. However, it should be noted that the ENSO variability is equator especially in the north tropical Pacific is too meridion- relatively weaker in the NEW model, apparent from the weaker ally confined. In addition, the wind stress anomalies from the height anomalies in the tropics in Fig. 18c. subtropics extend too far to the south. In contrast, the variability of windstress in the OLD model is unrealistic over the whole 5. Summary and discussion domain. In summary, the NEW COLA coupled climate model (COLA AGCMV3.2-MOM3) is comparable to many of the state of the 4.11. Relationship between Nino3˜ SST and precipitation art IPCC (AR4) models in its ENSO simulation (see Capotondi In Figs. 17aÐc, we show the regression of the Niño3 SST in- et al., 2006; Joseph and Nigam, 2006; Rao and Sperber, 2006). dex on the precipitation from observations (CMAP and SST is The climatological annual cycle of SST and wind stress in the from HADISST) and the OLD and the NEW model simula- eastern equatorial Pacific Ocean, the climatological annual mean tions, respectively. It should be noted that the reanalysis pre- zonal gradient of the thermocline along the equatorial Pacific cipitation is sensitive to the model convective parametrization Ocean, realistic evolution of ENSO percieved from subsurface scheme (Nigam and Chung, 2000) and therefore is not used here ocean variability, prevalence of neutral events, and the seasonal for validation. In Fig. 17a, we have used 22 yr of CMAP pre- phase locking of ENSO to the annual cycle are some of the major cipitation extending from 1979 to 2000 with corresponding SST improvements in the ENSO simulation of the NEW coupled from the HADISST dataset. climate model relative to the OLD model. The OLD model displays precipitation anomalies in the trop- This study also attempts to understand how such an improve- ical Oceans which is at variance with the observed anomalies. ment in the ENSO simulation occurred in the NEW model

Tellus 59A (2007), 3 304 V. MISRA ET AL.

Table 2. The centre of mass (C) of the regressed wind stress from

Fig. 16 calculated following Capotondi et al. (2006) as lx τ x (x)x dx C = 0 where, l is the width of the ocean, x is the lx τ x x 0 dx longitude and τ x is the anomalous zonal wind stress averaged between 2◦S and 2◦N at each longitude

NCEP reanalysis OLD NEW

164◦W Undeﬁned 176◦E

Additionally, we contend that intercomparing schemes in coupled climate models for the purposes of evaluating its efficacy in reproducing the observed mean and variable climate is futile as long as tunable parameters are present. However, based on our current understanding from several theoretical, observational and coupled modelling studies we pro- vide some physical explanations for the differences in the ENSO simulation between the OLD and the NEW model. The ENSO time scale increased to 3 yr (Fig. 7) in the NEW model primarily because of the broadening of the meridional scale of the zonal wind stress anomalies (Fig. 16) over the Central Pacific. Kirtman (1997) has shown that such broadening of the wind stress excites the slower off-equatorial Rossby waves which result in a longer time period for the phase reversal of the ENSO cycle. Another complimentary theory is that the zonal location of the zonal wind stress anomaly can play a significant role in the ENSO period (An and Wang, 2000). Following Capotondi et al. (2006) we have calculated the centre of mass of the regressed wind stress from Fig. 16 for the NCEP reanalysis, NEW and OLD model simulations (Table 2). It is seen from Table 2 that the NEW model’s centre of mass for the regressed wind stress is eastward and closer to NCEP reanalysis than the OLD model. In the OLD model the zonal wind stress anomalies are in fact outside the 2◦SÐ2◦N lat- Fig. 18. The regression of the DJF Niño3 SST index with itude band as a result of which the centre of mass is undefined. contemporaneous geopotential height (in meters) at 200 hPa from (a) Following the argument of An and Wang (2000) which suggests the NCEP reanalysis (b) the OLD and (c) the NEW coupled model that eastward shift of the zonal wind stress anomalies results in simulations. the increased contribution of the zonal feedback term in the SST tendency equation to growth of the ENSO cycle rather than its relative to the OLD. It is rather difficult to isolate a process for transition from one phase to the other also justifies the NEW the overall changes in the ENSO simulation of the NEW model model’s increased ENSO period. from the OLD model given that a number of changes were made The improvement in the seasonal phase locking of ENSO to the physical parametrization schemes and the numerics that arises from the improvement in the mean seasonal cycle of included changes to the vertical discretization and horizontal dif- the SST and the associated thermocline variability. Meinen and fusion to the AGCM of the NEW coupled model. Furthermore, McPhaden (2000) showed from subsurface observations that the Schneider (2002) showed that differences in the simulation over contemporaneous zonal tilt and the zonally averaged thermo- the eastern equatorial Pacific Ocean between two different cou- cline depth variations at quadrature with the SST variability in pled climate models could not be isolated to a dominant process the eastern Pacific are the two most dominant modes of the upper in their atmospheric components. However, he showed (as seen ocean water volume that dictate the evolution and sustenance in this study with identical ocean in the OLD and the NEW cou- of ENSO. Both these leading modes of the thermocline depth pled models) that the differences in the atmospheric models are (Figs. 10 and 11) are qualitatively similar in the NEW model and the prime drivers for the differences in the coupled simulations. ODA.

Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 305

the surface winds. A similar feature of strong meridional gradient of heating is seen in Fig. 17c. This pattern is nearly identical when precipitation anomalies are replaced with total diabatic heating in both the OLD and the NEW models (not shown). The more east-west orientation of the heating in the OLD model (Fig. 17b) is probably one of the few positive features of the ENSO simulation in the OLD model.

6. Conclusions In this study, we presented 12 salient observed features of ENSO to validate a coupled climate model for its ENSO simulation. Although, these features may not be exhaustive, they are suf- Fig. 19. The climatological area average (140◦EÐ80◦W and ficient to make a quantitative judgement on the veracity of the 10◦SÐ10◦N) annual mean total diabatic heating profile from the ENSO simulation, based on our present understanding of the (a) NEW and (b) the OLD coupled model simulations. phenomenon. A long term simulation of the newly developed COLA coupled Surface wind stress is an important source of thermocline climate model was verified for these 12 features and contrasted variability and in the maintenance of its mean state. A signifi- with an older version of the COLA coupled model. Overall, the cant improvement in the NEW model relative to the OLD model NEW coupled model is a major improvement over the OLD is seen in the mean annual cycle of the wind stress (Figs. 5 and model. A summary of the results is presented in Table 3. 6) and its variability (Fig. 16). Nigam et al. (2000) from their A fundamental difference, besides the differences in the ENSO linear diagnostic modelling study show that the spatial distribu- features, between the OLD and the NEW model simulations is tion and the vertical profile of diabatic heating have a profound the shift from the warm SST bias in the equatorial Pacific to a influence on the surface wind anomalies. They argue from their cold SST bias. This is of great relevance to many current state of diagnostic modelling study that a ‘bottom heavy’ heating profile the art coupled climate models that suffer from a cold bias and will entail stronger low level prevailing winds. The improvement concomitant easterly bias in the zonal wind stress over the equa- in the wind stress simulation of the NEW model is largely at- torial Pacific. This difference between the OLD and the NEW tributed to the change in the total diabatic heating profile shown in model cannot be isolated to a single physical process in the atmo- Fig. 19. In Fig. 19(a,b) it is clearly seen that the NEW (OLD) sphere, land or in the ocean. We contend that the improvement model heating profile is bottom (top) heavy. Furthermore, the seen in the total diabatic heating profile has significantly con- diabatic heating profile of the NEW model (Fig. 19a) is qualita- tributed to the overall improvement of the NEW model. If we tively and quantitatively quite similar to the corresponding pro- follow the argument of Nigam and Chung, (2000), who relate file of the ECMWF reanalysis (cf. fig. 8 of Nigam and Chung, the vertical heating distribution to bias in the surface wind stress 2000). and reconcile that surface wind stress is critical for the thermo- On the other hand Nigam and Chung, (2000) and Nigam and cline variability (without discounting the role of the net heat Chung, (2000) argue that the strong diabatic cooling in the off- and fresh water fluxes under weak wind conditions) in the deep equatorial tropics from a more Hadley-like heating redistribution tropics, then a clear relationship can be noticed in how diabatic (as in most coupled climate models) result in an easterly bias of processes are critical for the simulation of the tropical climate

Table 3. A short summary of comparison between the OLD and the NEW COLA coupled climate models

Feature OLD NEW

RMSE of SST in the Niño3 region 3.91◦ 0.94◦ Period of ENSO (determined from Niño3 Biennial 3 yr SST spectrum Seasonal phase locking of ENSO Inexistent Exists Duration of ENSO events 10 months 13 months Wind stress Erroneous (biennial feature Verifiable in zonal wind stress) Subsurface ocean Not analyzed Reasonable ENSO teleconnections Poor Relatively improved

Tellus 59A (2007), 3 306 V. MISRA ET AL.

(including its variability). However, it should be recognized that Collins, W. D., Hackney, J. K. and Edwards, D. P. 2002. An diabatic processes in the atmosphere (at least in the deep trop- updated parameterzation for infrared emission and absorption ics) is a result of coupled interactions and therefore cannot be by water vapor in the National Center for Atmospheric Re- isolated to a process in a single component model. An important search Community Atmosphere Model. J. Geophys. Res. 107(D22), realization of this study is that the development of the coupled doi:10.1029/200/JD0011365, 2002. Covey, C., Achuta Rao, K. M., Cubasch, U., Jones, P., Lambert, M. E. climate model should occur in a coupled framework and not in and co-authors. 2003. An overview of results from the coupled model the isolation of the component models. intercomparison project (CMIP). Global Planet. Change 37, 103Ð 133. Dirmeyer, P. A. and Zeng, F. J. 1999. Precipitation infiltration in the 7. Acknowledgments simplified SiB land surface scheme. J. Meteor. Soc. Japan 78, 291Ð The authors would also like to express their deep appreciation 303. to Drs. Byron Boville, William Collins of NCAR, Masao Kana- DeWeaver, E. and Nigam, S. 2004. On the forcing of ENSO teleconnections by anomalous heating and cooling. J. Climate 17, 3225Ð3235. mitsu of UCSD, Song-You Hong of Yonsei University, Shrinivas Enfield, D. B. and Mayer, D. A. 1997. Tropical atlantic sea surface Moorthi of NCEP, Julio Bacmeister of NASA and Ben Cash of temperature variability and its relation to El Niño southern oscillation. COLA for their invaluable suggestions at critical times of the de- J. Geophys. Res. 102, 929Ð945. velopment of the COLA AGCM. We would also like to acknowl- Gent, P. R. and McWilliams, J. C. 1990. Isopycnal mixing in ocean edge that comments from three anonymous reviewers helped us circulation models. J. Phys. Oceanography 25, 150Ð155. considerably in improving the content of this manuscript. This Goddard, L. and Dilley, M. 2005. El Niño: Catastrophe or Opportunity. study was supported by NSF grant ATM0332910, NASA grant J. Climate 18, 651Ð665. NNG04GG46G and NOAA grant NA04OAR4310034. The au- Gu, D. and Philander, S. G. H. 1995. Secular changes of annual and thors would also like to acknowledge the support of the com- interannual variability in the tropics during the past century. J. Climate puting infrastructure provided by the National Center for Atmo- 8, 864Ð876. spheric Research computing systems laboratory. Guilyiardi, E. 2006. El Niño mean state-seasonal cycle interactions in a multimodel ensemble. Clim. Dyn., 26, 329Ð348. Harshvardhan, R., Davies, R., Randall, D. A. and Corsetti, T. G. 1987. References A fast radiation parameterization for atmospheric circulation models. J. Geophys. Res. 92(D1), 1009Ð1016. An, S.-I. and Wang, B. 2000. Interdecadal change of the structure of the Holtslag, A. A. M. and Moeng, C.-H. 1991. Eddy diffusivity and coun- ENSO mode and its impact on the ENSO frequency. J. Climate 13, tergradient transport in the convective atmospheric boundary layer. J. 2044Ð2055. Atmos. Sci. 48, 1690Ð1698. An, S.-I. and Jin, F.-F. 2001. Collective role of thermocline and zonal Hong, S. Y. and Pan, H. L. 1996. Nonlocal boundary layer vertical dif- advective feedbacks in the ENSO mode. J. Climate 14, 3421Ð3432. fusion in a medium range forecast model. Mon. Wea. Rev. 124, 2322Ð Alexander, M. A., Blade, I., Newmann, M., Lazante, J. R., Lau, N.-C. 2339. and co-authors. 2002. The atmospheric bridge: the influence of enso Horel, J. D. and Wallace, J. M. 1981. Planetary-scale atmospheric phe- teleconnections on air-sea interaction over the global oceans. J. Cli- nomena associated with the Southern Oscillation. Mon. Wea. Rev. 109, mate 15, 2205Ð2231. 813Ð829. Bacmeister, J. T., Pegion, P. J., Schubert, S. D. and Suarez, M. J. 2000. Hoskins, B. J. and Karoly, D. J. 1981. The Steady Linear Response of a An atlas of seasonal means simulated by the NSIPP-1 atmospheric Spherical Atmosphere to Thermal and Orographic Forcing. J. Atmos. GCM. 17, NASA Tech. Memo-2000Ð104606, Goddard Space Flight Sci. 38 1179Ð1196. Center, Greenbelt, MD (194 pp). Huang, B. 2004. Remotely forced variability in the tropical Atlantic Bohren, C. F. and Albrecht, B. A. 1998. Atmospheric Thermodynamics. Ocean. Clim. Dyn. 23, 133Ð152, doi:10.1007/s00382-004-0043-8. Oxford University Press, New York, 402 pp. Ineson, S. and Davey, M. K. 1997. Interannual climate simulation and Bjerknes, J. 1969. Atmospheric teleconnections from the equatorial Pa- predictability in a coupled TOGA GCM. Mon. Wea. Rev. 125, 721Ð cific. Mon. Wea. Rev. 97, 820Ð829. 741. Briegleb, B. P. 1992. Delta-Eddington approximation for solar radiation International GEWEX Project Office 2002. The Second Global Soil Wet- in the NCAR Community Climate Model. J. Geophys. Res. 97, 7603Ð ness Project Science and Implementation Plan. IGPO Publication Se- 7612. ries No. 37, 69 pp. Capotondi, A., Wittenberg, A. and Masina, S. 2006. Spatial and temporal Jin, F.-F. 1997. An equatorial recharge paradigm for ENSO. Part I: con- structure of Tropical Pacific interannual variability in 20th century ceptual model. J. Atmos. Sci. 54, 811Ð829. coupled simulations. Ocean Modeling, 15, 274Ð298. Joseph, R. and Nigam, S. 2006. ENSO evolution and teleconnections in Chang, P., Saravanan, R. and Ji, L. 2003. Tropical Atlantic seasonal pre- IPCC’s 20th century climate simulations: realistic representation? J. dictability: the roles of El Niño remote influence and thermodynamic Climate 19, 4360Ð4377. air-sea feedback. Geophys. Res. Lett. 30, 1501Ð1504. Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D. and Chen, W. Y. 2004. Significant change of extratropical natural variability co-authors. 1996. The NCEP/NCAR 40-year reanalysis project. Bull. associated with tropical ENSO anomaly. J. Climate 17, 2019Ð2030. Amer. Soc. 77, 437Ð471.

Tellus 59A (2007), 3 VALIDATING AND UNDERSTANDING THE ENSO SIMULATION 307

Kiehl, J. T., Hack, J. J., Bonan, G., Boville, B. A., Williamson, D. L. Pacanowski, R. C. and Griffies, S. M. 1998. MOM3.0 manual, and co-authors. 1998. The National Center for Atmospheric Research NOAA/Geophysical Fluid Dynamics Laboratory, Princeton, New Community Climate Model: CCM3. J. Climate 11, 1131Ð1149. Jersey, USA 08542. Kirtman, B. P. 1997. Oceanic rossby wave dyanmics and the ENSO Philander, S. G. H. 1983. El Niño-Southern Oscillation phenomena. period in a coupled model. J. Climate 10, 1690Ð1704. Nature 302, 295Ð301. Kirtman, B. P. 2003. The COLA anomaly coupled model: ensemble Philander, S. G. H. 1990. El Nino,˜ La Nina˜ and the Southern Oscillation. ENSO prediction. Mon. Wea. Rev. 131, 2324Ð2341. Academic Press, San Diego, USA, pp. 293. Kirtman, B. P. and Shukla, J. 2002. Interactive coupled ensemble: a Philander, S. G. H. 2001. El Ninõ Southern Oscillation (ENSO) Models. new coupling strategy for GCMs. Geophys. Res. Lett. 29, 1029Ð In: Encyclopedia of Ocean Sciences (eds J. H. Steele, S. A. Thorpe, 1032. and K. K. Thurekian), Academic Press, London. Kirtman, B. P., Fan, Y. and Schneider, E. K. 2002. The COLA global Rao, K. A. and Sperber, K. 2006. ENSO simulation in coupled ocean- coupled and anomaly coupled ocean atmosphere GCM. J. Climate 15, atmosphere models: are the current models better? Clim. Dyn. 27, 2301Ð2320. 1Ð15, doi:10.1007/s00382-006-0119-7. Krishnamurthy, V. and Kirtman, B. P. 2003. Variability of the Indian Rasmusson, E. M. and Carpenter, T. H. 1982. Variations in Tropical Sea Ocean: relation to monsoon and ENSO. Quart. J. Roy. Meteor. Soc. Surface Temperature and Surface Wind Fields Associated with the 129, 1623Ð1646. Southern Oscillation/ El Niño. Mon. Wea. Rev. 110, 354Ð384. Kumar, A. and Hoerling, M. P. 1998. Annual cycle of Pacific-North Rayner, N. A., Parker, D. E., Horton, E. B., Folland, C. K., Alexander, L. American seasonal predictability associated with different phases of V. and co-authors. 2002. Global analyses of sea surface temperature, ENSO. J. Climate 11, 3295Ð3308. sea ice and night marine air temperature since the late nineteenth Large, W. G., McWilliams, J. C. and Doney, S. C. 1994. Oceanic ver- century. J. Geophys. Res. 108, doi:10.1029/2002JD002670. tical mixing: a review and a model with a nonlocal boundary layer Redi, M. H. 1982. Oceanic isopchynal mixing by coordinate rotation. J. parameterization. Rev. of Geophys. 32, 363Ð403. Phys. Oceanography 11, 1443Ð1451. Latif, M., Sperber, K., Arblaster, J., Braconnot, P., Chen, D. and Ropelewski, C. F. and Halpert, M. S. 1987. Global and Regional scale co-authors. 2001. ENSIP: The El Niño simulation intercomparison precipitation patterns associated with the El Niño/Southern Oscilla- project. Clim. Dyn. 18, 255Ð272. tion. Mon. Wea. Rev. 115, 1606Ð1626. Lau, N.-C. and Nath, M. J. 1996. The Role of the “AtmosphericBridge” in Rosati, A., Gudgel, R. and Miyakoda, K. 1997. The impact of ocean Linking Tropical Pacific ENSO Events to Extratropical SST Anoma- initial conditions on ENSO forecasting with a coupled model. Mon. lies. J. Climate 9, 2036Ð2057. Wea. Rev. 125, 754Ð772. Levitus, S. 1982. Climatological World Atlas of the Ocean. US Govt. Schneider, E. K. 2002. Understanding differences between the Equato- printing office. 173 pp. rial Pacific as simulated by two coupled GCMs. J. Climate 15, 449Ð Marx, L. 2002. New calculation of saturation specific humidity and sat- 469. uration vapor pressure in the COLA atmospheric general circulation Smagorinsky, J. 1963. General circulation experiments with the primitive model. Available from ftp://grads.iges.org/pub/ctr/ctr 130 text.pdf. equations: I. The basic experiment. Mon. Wea. Rev. 91, 99Ð164. Mechoso, C. R., Robertson, A. W., Barth, N., Davey, M. K., Delecluse, Straus, D. and Shukla, J. 2000. Distinguishing between the SST-forced P. and co-authors. 1995. The seasonal cycle over the torpical Pacific variability and internal variability in mid-latitudes: analysis of ob- in coupled Ocean-Atmosphere general circulation models. Mon. Wea. servations and GCM simulations. Quart. J. Roy. Meteor. Soc. 126, Rev. 123, 2825Ð2838. 2323Ð2350. Meehl, G. A. and Arblaster, J. M. 1998. The Asian-Australian Mon- Suarez, M. J. and Schopf, P. S. 1988. A delayed action oscillator for soon and El Niño Oscillation in the NCAR Climate System Model. J. ENSO. J. Atmos. Sci. 45, 549Ð566. Climate 11, 1356Ð1385. Tiedtke, M. 1984. The effect of penetrative cumulus convection on the Mellor, G. L. and Yamada, T. 1982. Development of a turbulence closure large-scale flow in a general circulation model. Beitr. Phys. Atmos. 57, model for geophysical fluid processes. Rev. Geophys. Space Phys. 20, 216Ð239. 851Ð875. Timmermann, A., 1999. Detecting the nonstationary response of ENSO Moorthi, S. and Suarez, M. J. 1992. Relaxed Arakawa-Schubert. A pa- to greenhouse warming. J. Atm. Sci. 56, 2313Ð2325. rameterization of Moist Convection for General Circulation Models. Trenberth, K. E., Branstator, G. W., Karoly, D., Kumar, A., Lau, N.-C., Mon. Wea. Rev. 120, 978Ð1002. and co-authors. 1998. Progress during TOGA in understanding and Meinen, C. S. and McPhaden, M. J. 2000. Observations of warm water modelling global teleconnections associated with tropical sea surface volume changes in the equatorial Pacific and their relationship to El temperatures. J. Geophys. Res. 103, 14 291Ð12 324. Niño and La Niña. J. Climate 13, 3551Ð3559. Troen, I. and Mahrt, L. 1986. A simple model of the atmospheric bound- Neelin, J. D., Latif, M., Allaart, M. A. F., Cane, M. A., Cubasch, U. and ary layer. Sensitivity to surface evaporation. Bound Layer Meteor. 37, co-authors. 1992. Tropical air-sea interaction in general circulation 129Ð148. models. Clim. Dyn. 7, 73Ð104. Tziperman, E., Zebiak, S. E. and Cane, M. A. 1997. Mechanisms of Nigam, S., Chung, C. and DeWeaver, E. 2000. ENSO diabatic heating in seasonal-ENSO interactions. J. Atmos. Sci. 54, 61Ð71. ECMWF and NCEP-NCAR reanalyses and NCAR CCM3 simulation. van Oldenborgh, G. J., Philip, S. and Collins, M. 2005. El Niño in a J. Climate 13, 3152Ð3171. changing climate: a multimodel study. Ocean Sci. 1, 81Ð95. Nigam, S. and Chung, C. 2000. ENSO surface winds in CCM3 simula- Wang, B. 1995. Interdecadal changes in El Niño onset in the last four tions: diagnosis of errors. J. Climate 13, 3172Ð3186. decades. J. Climate 8, 267Ð285.

Tellus 59A (2007), 3 308 V. MISRA ET AL.

Wu, R., Kirtman, B. P. and Pegion, K. 2006. Local Air-sea relationship to soil properties. A case study using HAPEX-Mobilhy data. Global in observations and model simulations. J. Climate, 19, 4914Ð4932. Planetary Change 13, 183Ð194. Xie, P. and Arkin, P. 1996. Analysis of global monthly precipitation Yang, X. Q. and Anderson, J. L. 2000. Correction of systematic errors using guage observations, satellite estimates and numerical model in coupled GCM forecasts. J. Climate 13, 2072Ð2085. predictions. J. Climate 9, 840Ð858. Zelle, H., Appeldoorn, G., Burgers, G. and van Oldenborgh, G. J. 2004. Xue, Y.-K., Sellers, P. J., Kinter, J. L. and Shukla, J. 1991. A simpliﬁed The relationship between sea surface temperature and thermocline biosphere model for global climate studies. J. Climate 4, 345Ð364. depth in the Eastern Equatorial Paciﬁc. J. Phys. Oceanography 34, Xue, Y.-K.,Zeng, F. J. and Schlosser, C. A. 1996. SSiB and its sensitivity 643Ð655.

Tellus 59A (2007), 3